Adaptive Grouping Sparse Bayesian Learning for Channel Estimation in Non-Stationary Uplink Massive MIMO Systems

Abstract

This paper addresses wideband channel estimation in an uplink massive multiple-input multiple-output (MIMO) system, which consists of multiple single-antenna users and a base station (BS) equipped with a large number of antennas. Due to the spatially non-stationary characteristics in massive MIMO channels, the channel vectors corresponding to different BS antennas may take on different sparsity patterns in delay domain. Taking this into account, we propose an adaptive grouping sparse Bayesian learning (AGSBL) scheme for uplink channel estimation. Specifically, for the channel vectors to be estimated, we judiciously design a hierarchical prior model controlled by a set of tunable hyperparameters. Then, we develop a variational Bayesian algorithm to infer the posterior probability of the channel vectors and the hyperparameters associated with the prior, thereby obtaining the channel estimates. Due to the designed prior, we are able to adaptively divide the channel vectors involved into several groups, each group bearing a similar sparsity pattern. In this way, the AGSBL can efficiently exploit the partial joint sparsity shared by each group of channel vectors, and therefore significantly improves the channel estimation performance. The simulation results are provided to verify the advantages of the proposed AGSBL over the related counterparts.